Ramp says AI spend is stalling. For buyers, that's the best news yet

Sounds like a crisis: corporate AI spending is stalling. It isn't. It's what happens in every market when the product turns into a commodity.

The data comes from Ramp, a US corporate card and spend management company that sees what its customers pay for. According to Ramp data reported by TechCrunch on November 6, token spend grew just 1.5% across August and September combined, even as volumes kept rising. Compare that with July, when spend jumped 42% month over month. October picked up again to 8.9%. Over the same stretch, token volume grew 37.1%, per the Ramp AI Index.

The reason is on the price sheets. In September, OpenAI halved API prices for GPT-6 Sol and GPT-6 Luna. Anthropic cut prices on Fable 5.1 and Opus 5.5 significantly and, according to TechCrunch, said in a filing that the cuts reduced its revenue growth by about $600 million.

This is a price war, not a demand problem

Ramp's lead economist Ara Kharazian puts it plainly: demand isn't weak, AI just got cheaper. At the same time, companies are adding cost controls. Expensive models are blocked by default for long tasks, employees get budgets, routine work gets routed to cheaper models.

Anyone who has bought media knows this pattern. First you pay for reach, then you pay for results. AI procurement is making that exact move right now. The question is no longer “which model is best?” but “what does one finished task cost me?” Model routing is the new media mix.

Who wins, who pays

The way I see it, anyone buying right now wins. The two big vendors are fighting for share: per Ramp, Anthropic took 60.5% of token spend in the week to October 23, OpenAI 35.1%. In a market like that, discounts are negotiable and long commitments are a bad deal.

The spread is telling, too. Median spend is $12 to $15 per employee per month, while the top 10% pay $600 to $1,000. Kharazian says almost all of the spend comes from venture-backed tech companies. The typical mid-sized business is nowhere near the cost trap that heavy users are discovering now.

The catch

Ramp measures US companies paying through Ramp, skewed toward tech. For Europe, these are signals, not market data. And cheaper tokens don't automatically mean cheaper projects: run more agents and you'll burn many times the volume.

What I'd do now

• No annual contracts at today's prices without a price adjustment clause. Prices are falling faster than contract terms run.
• Measure cost per task, not per seat. Pick three typical tasks and price them out on two models.
• Set a simple routing rule: a cheap model for routine work, the expensive one only where it makes the difference.

Bottom line: Once AI becomes a commodity, the winner isn't the biggest user. It's the best buyer.